An Intelligent and Scalable Hybrid Deep Learning Architecture for Real Time Human Activity Recognition in Maritime Operations

Abstract

The growing integration of different intelligent monitoring systems in maritime environments has created a strong demand and necessity to automate methods for recognizing and interpret human activities in real time. Human Activity Recognition (HAR) plays an important role to enhance safety, operational efficiency, and situational awareness in maritime operations such as navigation support, rescue missions, maintenance tasks, and crew monitoring etc. This research proposes a scalable hybrid deep learning architecture for real-time human activity recognition in maritime operations by integrating both sensor-based and vision-based data sources. This framework uses data collected from smart phones, wearable and smart devices equipped with embedded sensors such as accelerometers and gyroscopes, along with wireless connectivity modules such as Bluetooth, combined with digital image and video streams captured in maritime environments. Two models are developed: a sensor-based model architecture and a video-based model architecture, employing Ensemble Convolutional Neural Networks (ECNN) integrated with a Machine Learning Voting Classifier (MLVC) to improve activity recognition performance. The proposed architecture is designed to support scalable deployment and real-time activity monitoring. Experimental evaluation is conducted using recognition accuracy and comparative analysis with existing state-of-the-art approaches. Results demonstrate that the hybrid multimodal framework significantly improves classification accuracy, robustness, and real-time performance, making it suitable for practical deployment in intelligent maritime monitoring systems.